SIMBAD references

Gamma-ray bursts (GRBs) with extended emission (EE) challenge traditional GRB classification schemes, yet their origins remain unknown. In this paper, we employ unsupervised machine learning methods to classify Swift EE GRBs using multidimensional observational parameters. We apply PCA-tSNE and PCA-UMAP for dimensionality reduction followed by AutoGMM clustering and hierarchical consolidation, yielding two distinct machine-classified populations (Merged Clusters I/II). These clusters are systematically evaluated against established classification frameworks, including the Amati-relation-based classification, the Dainotti fundamental plane correlation, the duration-based long/short GRB taxonomy, and the EE GRB categories associated with kilonovae/supernovae (KNe/SNe). Our results indicate that Merged Cluster I (EE-I) predominantly exhibits an association with long GRBs, corresponding to progenitors from massive stellar collapse, while Merged Cluster II (EE-II) is primarily composed of short GRBs, with characteristics consistent with compact binary merger events. The clusters' distinct properties are further validated by the Dainotti relation, with Cluster II showing a tighter fundamental plane correlation. Moreover, the concentration of plateau shift outliers within this cluster reveals properties consistent with a potential merger origin. While the traditional T₉₀ remains broadly effective for subclass discrimination, we demonstrate critical discrepancies that expose inherent limitations of univariate classification. Additionally, our results show that SN/KN-associated EE GRBs typically appear in separate subclusters. Crucially, the discovery of hybrid GRBs unexpectedly coexisting within a shared subcluster reveals remarkable complexity and diversity in the physical origins of EE GRBs. We hope that this data-driven machine learning approach can provide meaningful insights into the physical origins and subclass differentiation of EE GRBs.